Electronic Warfare Associates
Signal analysis for models at the edge
- 600kIQ samples processed, 10M complex values each
- 99,999spectrogram images, processed within edge-hardware limits
I architected spectrogram-based frequency-domain analysis for an enhanced edge model intended for military zones, processing a dataset of 99,999 images efficiently while staying within the target hardware’s constraints.
I engineered a custom CNN variational autoencoder in Keras to synthesize smooth signal manifolds, and processed a large-scale IQ dataset of 600,000 samples with 10 million complex values each.